EXP-011 verdict — Turtle + Bollinger squeeze, 30-symbol generalization
- Run date: 2026-09-10
- Execution model:
claude-opus-5 / high - Verdict:
rejected - 2026-09-10 correction: the first verdict preserved the drawdown reduction as
component-hold, but this was withdrawn (see below)
| generated_at | 2026-09-10T00:00:00+09:00 |
Grounds for the verdict
Of the 8 preregistered promotion gates, only 5, 6, 7 and 8 passed (direction held under stress, direction held across neighbouring parameters, consistent direction across periods, 0 capital starvation in the verdict cell). But these mean that negative results reproduce consistently, so they are not grounds for promotion.
| Gate | Criterion | Result | Pass |
|---|---|---|---|
| 1. Median CAGR of the 23 new symbols | ≥ 15% | -1.70% (turtle_squeeze_60) | ✗ |
| 2. Median MDD | ≥ -25% | -15.14% | ✓ |
| 3. Median excess CAGR | > 0 | -19.03% | ✗ |
| 4. Total locked OOS trades | ≥ 100 | 89 | ✗ |
| 5. Direction held at 2x costs and 5bp | Held | Held (negative) | ✓ |
| 6. Direction held at neighbouring ATR 1.0 and 2.0 | Held | Held (negative) | ✓ |
| 7. Consistent across train, validation and OOS | Consistent | Consistent (negative in every period) | ✓ |
| 8. Verdict cell starved | 0 | 0 | ✓ |
Gates 1, 3 and 4 were not passed, so it is not advance. Furthermore, because the effect existed only in the 7 control symbols and disappeared in the 23 new symbols, it is recorded as rejected, not hold. hold means deferring judgment for lack of samples; this time, judgment was reached by enlarging the sample.
What was rejected and what remained
Rejected: the path of promoting Turtle + Bollinger squeeze as a profit strategy. EXP-008's CAGR of 9.33% was an artefact of 7 symbols; on the 23 new symbols it is -1.70%. The median is negative even at 0bp commission, so there is no room to revive it through cost normalization either. This contract's parameters are not explored further.
Preserved: the squeeze filter's drawdown reduction effect. Control -35.29% → -16.56%, new 23 symbols -32.27% → -15.14%, reproduced at all three ATR multiples. Since this was the only effect that held when the symbol basket changed, it is kept in component-hold status not as a strategy but as a risk-component candidate that can be attached to other entry rules . It is not grounds for standalone promotion; combined validation follows only after an entry rule with positive expectancy is secured first.
Side checks
Even after switching to per-period capital isolation accounting, the EXP-008 control figures reproduced (turtle_squeeze_60 9.74% vs 9.33%, MDD -16.56% vs -16.30%). So EXP-008's conclusion was not distorted by capital carry-over. The capital carry-over problem did real damage in the 5-minute contracts (EXP-006/009) and did not change verdicts in the daily contracts.
Putting the buy-and-hold benchmark in the table for the first time showed that the 30-symbol median in the locked OOS period was 18.88%. This means that even a positive CAGR in this period would most likely have merely reconfirmed the market's rise, and from now on every experiment records excess CAGR as well.
Not carried into the next verdict
- Further parameter search on the squeeze contracts
- Porting this contract to 5-minute or other resolutions
- Expansion to a small/mid-cap basket (there is no prior hypothesis that would overturn the current grounds for rejection)
Detailed results are in the 0 , and the design is in the 1 .
Correction — component-hold withdrawn (2026-09-10)
The first verdict recorded that "the squeeze filter's drawdown reduction reproduces when symbols change, so it is kept as a risk-component candidate to combine with other entry rules". This is withdrawn.
Measuring market exposure in the full-universe sweep showed the drawdown reduction was not a property of the filter but a result of non-participation. On the 30-symbol verdict basket, turtle_only exposure is 23.39%, while the squeeze contracts are at 3~5%. Normalized for exposure, the squeeze side is actually worse. Across all symbols, 26% had 0 trades, with an average of 1.46 trades per symbol over 3.65 years.
Drawdown lowered by holding cash is not a reusable component, so nothing in this contract is preserved. The verdict is rejected alone, and no further combined validation of this strategy family is started.
Details are in the 0 , and the full-universe cross-section is in 1 .
Source and migration evidence
Source: reports/verdicts/EXP-011-turtle-squeeze-basket.md
Original SHA-256: 83bce770f985ccefb61f13e464bcd4c830b9cdc81be3ec397cef773b3fc5570d
Raw data and account information are not part of the public package. The original results do not replace statistical or scientific revalidation. The English text is a translation; numbers, tables and charts are carried over unchanged.